Hi Elisabetta, although I am not sure on which metrics you are interested in, for the purpose of our group's research we usually use either CKJM (http://www.spinellis.gr/sw/ckjm/) or Percerons (http://percerons.com/index.php)

Hi Apostolos, basically I am interested in various metrics but I would like to use tools that are not language specific. 

For instance, CKJM is for Java. I have to check Percerons.

You said that you use those tools. Have you implemented tests to verify that the results they produce are correct?

That's really an issue. I have seen dozens of research papers (include mine) published in important venues having metrics collected by several different tools. Any of such studies have considered whether or not metric tools provided correct results. I would say that it is an open research question to be investigated scientifically.

Not sure if this is of help or not.

At a previous job, I maintained an internal version of a lines of code counter.  This counter could handle multiple languages (we wrote the flex tokenizer for each new language).  This would then be run on the source tree and generate the total number of lines, lines of comments, code, and blank lines.

We then integrated it with our source code control system and we could determine from revision to revision how much churn there was in a given source file.  Number of new lines, number of deleted lines, and number of changed lines.  It was extended to handle a range of versions (or even from date a to date b).  This was then used to correlate the various source files vs the customer found problems - and help identify modules that were considered the "riskiest" as they had the most changes/churn during a given time frame.

Pygount (https://pypi.org/project/pygount/) is slower but has support for many more languages that the usual suspects.

I use sourceMeter (https://www.sourcemeter.com/).

It supports Java, Javascript, C, C++, C# and Python. However, some measures are not collected from some languages: for instance, Halstead metrics are not collected from C and C++ code.

Concerning the validation of the measurement, this is a rather complex matter.

First of all, there are measures, such as the "lines of code" that can be defined in several different ways. For instance, LOC may possibly exclude comment lines, blank lines, lines containing only syntactic elements (e.g., "{" in C); even so, you still have to decide if a statement that spans two lines should be counted as one or two LOCs.

In such cases, what is really important is that the tool provides a reasonable set of measure implementations. For instance, sourceMeter yields lines of code, logical lines of code, and the number of statements.

In many cases, all you need is that the set of measures you use are all collected via the same tool, so that they your analysis is internally consistent.

Concerning other measures, I checked some, thus I am fairly sure that sourceMeter computes McCabe complexity correctly.

Other measures, like the coupling between object, I did not check. I just trust the tool. Also because any small error would not affect the analysis I carry out.